A large, paired dataset of robotic and handheld lumbar spine ultrasound with ground truth CT benchmarking.

Musculoskeletal disorders present significant socio-economic challenges globally, requiring innovative diagnostic and therapeutic approaches. Current intraoperative imaging techniques, including computed tomography (CT) and radiography, involve high radiation exposure and limited soft tissue visualization. Ultrasound (US) offers a non-invasive, real-time alternative but is underutilized intraoperatively, as it is highly observer-dependent. US enhanced by artificial intelligence shows high potential for observer-independent surgical guidance, decision support, and robot-assisted applications in orthopedics. Given the limited availability of high-quality, in vivo ultrasound data, we introduce a comprehensive dataset from a collection of handheld 2D US (HUS) and robot-assisted US (RUS) lumbar spine imaging in 63 healthy volunteers. This dataset includes demographic data, HUS and RUS imaging with synchronized 3D positioning data of the US probe, corresponding CT data, and CT-reconstructed 3D models of the lumbar vertebrae L1 to L5, establishing a robust baseline for machine learning algorithms. This extensive collection is the first anatomy dataset for the lumbar spine that includes paired CT, HUS, and RUS imaging, supporting advancements in intraoperative imaging and robot-assisted surgery.